Link Travel Time Prediction Based On Neural Network Research

The transportation question is a worldwide problem which troubled modern big cities. With the rapid development of urban construction in our country, private car ownership has increased dramatically. Experts predict that by 2020, China's private car ownership will increase to 100 millions, the traffic problem has currently become China's major problem faced each big or media-sized city. The establishment of intelligent transportation system (ITS) is an effective way to solve traffic problems. Transport induction is an important part of ITS, and the vehicle road travel time as one of important indicators which reflect traffic conditions is an important content of traffic induction. At present, the research about road travel time prediction at home and abroad has been very comprehensive. However, in view of the specific conditions about most of the domestic traffic detection equipment, we can only predict the average speed of road travel, so as to predict road travel time.This thesis, in the light of the application of traffic induction demand, makes analysis and research on the methods of the road travel time prediction. It discusses several typical road travel time forecasting models, analyses strengths and deficiencies of the current road travel time prediction models. Combined with the actual testing situation of the traffic flow in Kunming city, this paper determine to apply the traffic measured parameters of the traffic flow, share and average travel speed obtained by road coil ring sensor detector, as the basis of the data of analysis and research by the weighed average pretreatment. On this basis, this paper designs BP network and RBF network——models of artificial neural network prediction, and tries to use genetic algorithm optimization of BP network and particle swarm optimization of BP network models to predict the road travel time.In this thesis, based on measured traffic flow data, using MATLAB toolbox and the four neural network models which this paper designs and builds, carries out simulation of Link Travel Time Prediction in Xuefu road of Kunming. In the input layer, data are integrated and processed, and Four models are joined the average share of the entire road in the input end, taking five-minute interval, the output is the average speed of road travel. This paper provides the simulation results and makes comparison and analysis of these results. The results indicate that use genetic algorithm optimization of BP network and particle swarm optimization of BP network models to predict road travel time is feasible, but slightly lower than the RBF model; RBF neural network model which is used to predict road travel time has a better adaptability, accuracy and real-time, thus further proves that based on RBF (Radial Basis Function) neural network to predict link travel time is feasible and effective.